Anthropic vs. The Pentagon

Here's what we know—and what's leaking out.

In January, U.S. forces reportedly used Claude during the operation to capture Venezuelan President Nicolás Maduro. The AI was deployed through Palantir, which has deep DoD contracts. A senior Anthropic exec reached out to Palantir asking whether their model was involved. Palantir flagged the call to the Pentagon because—and this is a direct quote from a defense official—"any company that would jeopardize the operational success of our warfighters in the field is one we need to reevaluate our partnership with."

That's when it got real.

Defense Secretary Pete Hegseth is now reportedly close to designating Anthropic a "supply chain risk"—a label usually reserved for foreign adversaries like Huawei. That would force every military contractor to cut ties with Anthropic. The contract on the table is worth up to $200M, which is pocket change against Anthropic's $14B annual revenue. But here's the part that should make you sit up: Claude is supposedly the only AI model currently running inside the military's classified systems. The government doesn't exactly publish its tech stack.

And that's kind of the point. This whole story is being pieced together from leaks, anonymous officials, and reporting that keeps getting confirmed weeks after the fact. What's actually happening behind closed doors is almost certainly more than what's making it to the press. If the Pentagon is this public about squeezing Anthropic, imagine the conversations we're not hearing. These AI models are going to get pressed for every ounce of capability the government can extract—if that isn't already happening.

The positions: Anthropic isn't saying "no military use." They've said they're willing to loosen their terms. What they won't agree to is mass surveillance of Americans or fully autonomous weapons with no human in the loop. The Pentagon's counter: they want AI tools available for "all lawful purposes." No carve-outs. No asterisks.

Then came the security tool. On February 20, Anthropic launched Claude Code Security—an AI vulnerability scanner that reads code the way a human security researcher would. In its testing with Opus 4.6, it found over 500 vulnerabilities in production open-source codebases that had gone undetected for decades. The market's response was immediate. CrowdStrike dropped 8%. Cloudflare fell 8.1%. Okta slid 9.2%. JFrog got hammered 25%. The Global X Cybersecurity ETF closed at its lowest since November 2023.

Barclays called the selloff "illogical" and said Claude Code Security doesn't directly compete with any established cybersecurity business they cover. They're probably right in the short term. But investors aren't trading on today—they're pricing in a future where an AI lab can casually obsolete a chunk of the security stack as a side feature. The iShares tech software ETF is down 23% since January and heading for its largest quarterly decline since 2008. That's not a correction. That's a reckoning.

Where we land: Anthropic is fighting on two fronts—one over ethics, one over market disruption—and both fights are really about the same question. What happens when the most capable AI doesn't play by the old rules? The Pentagon standoff is setting the precedent for every AI lab's relationship with the military. The security tool selloff is setting the precedent for every AI lab's relationship with... everyone else. Neither fight has a clean ending.

Cody Schneider is the founder of Graphed and one of those people who's been quietly running circles around entire marketing departments by himself. He's a growth engineer—part marketer, part builder—and he's been using Claude Code as his one-person army for months.

He came on our show and walked through exactly how he did it—40 Facebook ads, 100 landing pages, 3 guest blog posts, 4 podcast bookings, 5 help desk articles, 25 tweets, and 2 software products. In a single day. No team. Just Claude Code and a stack of APIs.

Here's the full loop, step by step:

1. Set up your war room. Create a project folder with a .env file containing your API keys (ad platforms, CMS, email tools). This is your base camp. Claude Code reads from this folder and connects to everything you need.

2. Mine your customer's language. Before you build anything, research. Cody pulls Reddit threads where people describe their pain in their own words—not marketing speak, not persona docs. The exact phrases real humans use when they're frustrated. This becomes your ad copy and landing page language.

3. Build a bulk ad generator. Using Claude Code, Cody builds React components that render ad creatives, then converts them to PNGs via HTML-to-Canvas. The whole thing takes about 20 minutes to set up. Then you generate dozens of variations—different headlines, different hooks, different visuals—all from one system.

4. Test cheap, kill fast. Spend $100 over three days running every ad variation. Sort by CPC. The winners reveal which angles resonate. Kill everything else. No emotional attachment. The data decides.

5. Generate landing pages at scale. Take your winning ad angles and pipe them into auto-generated landing pages via Strapi CMS. Each page matches the language and hook of the ad that sends traffic to it. Claude Code handles the page creation—you just point it at the winning angles.

6. Measure everything. Build a dashboard tracking performance across every asset. The bottleneck isn't production anymore—Claude Code handles that in minutes. The new bottleneck is analysis. Knowing what's working, what to kill, and where to double down. That's where the human comes in.

Cody's running this playbook every day at Graphed, and you don't need to know how to code to do it. We turned the whole thing into a step-by-step playbook with the exact prompts and tool stack.

This guide to Claude Code for beginners went viral this week — 20+ projects built without writing a single line of code, and a 24-minute video with the exact playbook—from first install to running multiple AI agents in parallel. If you've been curious about Claude Code but figured it was "for developers," this is your on-ramp. Watch it free

Spotify's best devs haven't written code since December — Co-CEO Gustav Söderström said it on their Q4 earnings call. They use an internal AI platform called Honk, built on Claude Code. Engineers describe fixes in plain English from Slack on their phones and get a new app version back before they finish their commute. 50+ features shipped in 2025 this way. TechCrunch

Claude in PowerPoint is live — Not "AI generates an image of a slide." Native editable PowerPoint elements—real shapes, charts, diagrams—that respect your existing templates. Three workflows: blank-deck builder, pinpoint editor, bullet-to-visual converter. All paid Claude plans. Try it

HubSpot just bought Starter Story — Pat Walls built a bootstrapped media brand to 800K YouTube subscribers, a 275K newsletter, and 1.6M total audience. HubSpot acquired it to feed their distribution engine. Greg Isenberg's take: "media is becoming the base layer of software companies. As infrastructure, not as marketing." Distribution is the last moat.

Kipp bodnar (HubSpot CMO) on AEO — If you missed last week's interview, Kipp brought receipts on why your Google traffic has an expiration date. HubSpot grew AEO-driven traffic 15x in a year. 58% of Google searches now end in zero clicks. Worth the watch if you haven't seen it. Full interview

Google dropped Gemini 3.1 Pro — 77.1% on ARC-AGI-2, more than double its predecessor. The focus is agentic reasoning and multi-step workflows. Google quietly retook the frontier model lead this week and almost nobody noticed. Details